Wake word setup
The wake word lets you drive dictation without touching the keyboard. InkSpoke keeps a small, fully-offline listener running in the background; when it hears the name you chose, it starts a dictation for you. This page covers turning it on, picking your wake name, and tuning how — and when — it listens.
What the wake word does
Once enabled, InkSpoke listens for three spoken phrases, all built from a single name (default Annika):
- Wake — say the name ("annika") to start a dictation. The listening overlay opens just like a hotkey press.
- Finish — say "thank you annika" to stop and process what you said. InkSpoke transcribes, refines (if enabled), and injects the text at your cursor.
- Cancel — say "cancel annika" to throw the recording away without transcribing.
All three are case-insensitive, and the finish and cancel phrases are derived automatically from whatever name you pick.
The listener runs on-device — audio for wake detection never leaves your machine.
The wake-word engine (Vosk) recognizes English only. Names and phrases are matched against an English speech model; there's no non-English wake word today.
Turn it on
- Open Settings → Hands-Free → Wake Word.
- Toggle Enable Wake Word.
- The first time you enable it, InkSpoke asks to download the offline wake-word model (about 42 MB) and shows a consent dialog. Approve it, and the download runs once.
- That's it — say "annika" to start your first hands-free dictation.
┌─ Settings › Hands-Free ─────────────────────────────────────┐
│ [ Overview ] [ Wake Word ] [ Voice Commands PRO ] │
├─────────────────────────────────────────────────────────────┤
│ Enable Wake Word .................................... (●) │
│ │
│ Wake name [ Annika ▾ ] Prefix [ (none) ▾ ] │
│ ▶ Preview │
│ │
│ Say to start "annika" │
│ Say to finish "thank you annika" │
│ Say to cancel "cancel annika" │
│ │
│ Detection sensitivity ............... [ Very High ▾ ] │
│ Schedule (active hours) ............. ( ) [ Edit ] │
│ Idle auto-disable ................... ( ) [ 10 min ] │
│ Sleep auto-disable .................. (●) │
│ Pause on music playback ............. (●) │
│ Continuous recording after wake ..... ( ) │
│ Quick confirm ....................... (●) │
└─────────────────────────────────────────────────────────────┘
The Overview tab (the first sub-tab) gives you an at-a-glance summary of everything InkSpoke is currently listening for, if you'd rather start there.
Pick your wake name
InkSpoke ships with a curated list of names that are reliably recognized by the offline model. Choose one from the Wake name picker, then click Apply (it appears only once you've changed something).
| Name | Name |
|---|---|
| Annika (default) | Samantha |
| Athena | Serena |
| Cassandra | Stella |
| Miranda | Victoria |
| Rebecca | Zeno |
Preview speaks the wake phrase aloud using your operating system's text-to-speech voice, so you can hear how it sounds before committing. (If your system has no TTS voice available, the preview is simply skipped.)
Add a greeting prefix
A single short name can occasionally misfire. Adding an optional greeting prefix gives the listener a longer, more distinctive phrase to match:
| Prefix | Start phrase becomes |
|---|---|
| (none) (default) | "annika" |
| Hey | "hey annika" |
| Hi | "hi annika" |
The finish ("thank you annika") and cancel ("cancel annika") phrases stay the same — only the trigger to start gets the prefix.
Pro and Perpetual users can pick Custom… in the name picker and type any name they like. As you type, InkSpoke checks it live (✓ Available / ✗ Not found). If the word already exists in the offline vocabulary it works instantly; if not, InkSpoke asks the InkSpoke servers to build a custom voice model for that name, downloads it, and merges it locally — this takes a moment and needs a one-time larger model download. It's still English-only. On a downgrade to Free, a custom name reverts to Annika.
Detection sensitivity
Sensitivity controls how eagerly the listener accepts what it hears as your wake word. Higher sensitivity catches more attempts (you miss fewer wakes) but is more prone to false triggers; lower sensitivity is stricter.
| Level | Behavior |
|---|---|
| Low | Strictest — fewest false triggers, but you may need to repeat the name |
| Normal | Balanced |
| High | Eager |
| Very High (default) | Most eager — triggers most readily, at the cost of occasional false positives |
If InkSpoke keeps waking up on its own during conversations, drop the sensitivity to High or Normal. If it's ignoring you, nudge it up. A prefix (above) also helps a lot.
Choose when it listens
An always-on listener doesn't have to be always-on. These controls scope when the wake word is active so it isn't burning CPU — or catching stray words — while you're away, asleep, or playing music.
| Setting | Default | What it does |
|---|---|---|
| Schedule (active hours) | Off (listens 24/7) | Restrict listening to time windows you set per day (e.g. Mon–Fri 08:00–18:00, weekends off). Edit opens a per-day editor; overnight ranges like 22:00–06:00 are supported |
| Idle auto-disable | Off | Stop listening after a stretch of no keyboard or mouse activity, and resume when you're back. Default timeout 10 minutes (adjustable 1–120) |
| Sleep auto-disable | On | Stop listening when the computer sleeps or the screen locks |
| Pause on music playback | On | Pause the listener when media is playing and that audio is bleeding into your mic — see the note below |
This setting doesn't touch Spotify, Apple Music, or your browser. When your system reports that media is playing and InkSpoke detects that sound leaking into the microphone, it quietly pauses the wake-word listener so a song lyric can't trigger it — then resumes when the music stops. If you're on headphones (nothing bleeds into the mic), nothing is paused.
Two ways the dictation flows
By default, saying the wake name just opens the overlay — you then speak your dictation. Two options change that flow:
| Setting | Default | What it does |
|---|---|---|
| Continuous recording after wake | Off | Start capturing in the same breath as the wake phrase — say "annika, send this email…" in one go. InkSpoke seeds a fraction of a second of audio from just before the wake and strips the wake phrase itself from the final transcript |
| Quick confirm | On | After your text is injected, you get a short window to say a confirm phrase — default "ok confirm" — to press Enter for you. Spoken without the wake name; "ok" and "okay" both count. Default window 60 seconds (adjustable 1–300) |
If you use a cloud speech model, you may prefer to leave Continuous recording after wake off — with it on, the fraction of a second captured before you finished the wake phrase is part of the audio that gets uploaded.
The Hands-Free toggle hotkey
You can turn wake-word listening on and off from the keyboard without opening Settings. This is handy when you're heading into a call or a noisy room and want the listener to stand down for a bit.
| Action | Windows / Linux | macOS |
|---|---|---|
| Toggle Hands-Free (wake word) on/off | Alt + Shift + H | ⌥ + ⇧ + H |
Like every InkSpoke shortcut, this one is configurable in Settings → General & hotkeys.
Where to go next
Wake word gets you hands-free dictation. If you want to open apps, press keys, or run whole actions by voice — "annika, action open Slack" — that's Voice Commands, a Pro feature that builds on the same listener. And teaching InkSpoke the names of your apps makes those commands far more reliable.
Next steps
- Voice commands — map spoken phrases to actions like opening apps, pressing keys, and inserting text (Pro).
- Favorites and vocabulary — add your apps and domain words so the offline listener recognizes them.
- Settings → General & hotkeys — rebind the Hands-Free toggle and every other shortcut.
- The listening overlay — what you see once a wake word starts a dictation.